Meta has released Muse Glimmer, a new language model with 30 billion parameters, distributed under the permissive Apache 2.0 license. This marks a return to openly available weights from Meta and removes some of the licensing friction associated with earlier Llama releases.
Technical details and local usability
The compressed model file is about 18.16 GB, which makes local deployment realistic for many users. According to shared observations, machines with at least 32 GB of RAM can run the model while still leaving room for other applications; the commenter noted their own system has 128 GB of RAM.
Why this matters
The Apache 2.0 license simplifies development and commercial use compared with more restrictive or ambiguous licenses. At 30 billion parameters, Muse Glimmer occupies a middle ground: it offers greater capacity than smaller models with only a few billion parameters, while remaining far less resource-intensive than models with hundreds of billions of parameters.
Community reaction and availability
The release and early user experiences were discussed on platforms such as Hacker News. Because of its relatively compact file size and permissive license, Muse Glimmer is a candidate for local experimentation, research projects, and deployments where an open license is important.
Summary
Muse Glimmer is Meta's 30B-parameter model released under Apache 2.0, with a compressed size of 18.16 GB. Its license and moderate resource requirements make it appealing for developers and researchers looking to run models locally or integrate them into applications.
